Methods and systems for identifying a re-routed transaction

Abstract

Embodiments provide methods and systems for identifying a re-routed transaction. Method performed by processor includes retrieving a plurality of transaction windows from a transaction database. Each transaction window includes a transaction declined under a restricted MCC. The method includes accessing a plurality of features associated with each transaction of each transaction window from the transaction database. The method includes predicting an output dataset of a plurality of reconstructed transaction windows based on feeding the input dataset to a trained neural network model. The method includes computing a corresponding reconstruction loss value for each transaction of each transaction window. The method includes comparing the corresponding reconstruction loss value for each transaction with a pre-determined threshold value. The method includes identifying the transaction as a re-routed transaction corresponding to the transaction declined under the restricted MCC, if a corresponding reconstruction loss value for a transaction is higher than the pre-determined threshold value.

What is claimed is: 
     
         1 . A computer-implemented method comprising:
 retrieving, by a processor, a plurality of transaction windows from a transaction database, each transaction window of the plurality of transaction windows comprising a transaction declined under a restricted Merchant Category Code (MCC) and each transaction window associated with a respective payment instrument;   accessing, by the processor, a plurality of features associated with each transaction of each transaction window from the transaction database, the plurality of features comprising a sequential merchant similarity value with respect to a previously generated sequential restricted MCC merchant embedding, a graph based merchant similarity value with respect to a previously generated graph based restricted MCC merchant embedding, a sequential card similarity value with respect to a previously generated sequential risky card embedding, a graph based card similarity value with respect to a previously generated graph based risky card embedding, and a sequential merchant MCC similarity value with respect to a previously generated sequential restricted MCC embedding;   generating, by the processor, an input dataset by appending the plurality of features to each associated transaction for each transaction of each transaction window of the plurality of transaction windows;   predicting, by the processor, an output dataset comprising a plurality of reconstructed transaction windows based on feeding the input dataset to a trained neural network model;   computing, by the processor, a corresponding reconstruction loss value for each transaction of each transaction window wherein a reconstruction loss value for an individual transaction is computed by comparing the individual transaction with a corresponding reconstructed transaction;   comparing, by the processor, the corresponding reconstruction loss value for each transaction of each transaction window with a pre-determined threshold value; and   if a corresponding reconstruction loss value for a transaction of a transaction window is higher than the pre-determined threshold value, identifying, by the processor, the transaction as a re-routed transaction corresponding to the transaction declined under the restricted MCC of the transaction window.   
     
     
         2 . The method as claimed in  claim 1 , wherein the restricted Merchant Category Code (MCC) is MCC 7995. 
     
     
         3 . The method as claimed in  claim 1 , wherein the sequential restricted MCC merchant embedding is previously generated for training the neural network model by computing a weighted average of a plurality of sequential restricted MCC merchant embeddings generated using a sequence model, and wherein the sequence model is trained to generate corresponding sequential merchant embeddings for all merchants retrieved from the transaction database. 
     
     
         4 . The method as claimed in  claim 3 , further comprising:
 generating corresponding sequential card embeddings for all cards retrieved from the transaction database using a weighted average of the sequential merchant embeddings;   identifying a plurality of risky cards having performed a higher number of restricted MCC transactions compared to a predefined restricted MCC threshold value; and   generating the sequential risky card embedding for training the neural network model by computing an average of a plurality of sequential risky card embeddings associated with the plurality of risky cards.   
     
     
         5 . The method as claimed in  claim 3 , wherein the sequential restricted MCC embedding is previously generated for training the neural network model using a sequence model, and wherein the sequence model is trained to generate corresponding sequential MCC embeddings for all MCCs retrieved from the transaction database. 
     
     
         6 . The method as claimed in  claim 1 , wherein the graph based restricted MCC merchant embedding is previously generated for training the neural network model by computing a weighted average of a plurality of graph based restricted MCC merchant embeddings generated using a Heterogenous Graph Convolutional Network (HGCN) model, and wherein the HGCN model is trained to generate corresponding graph based merchant embeddings for all merchants retrieved from the transaction database and wherein the HGCN model is trained to simultaneously generate corresponding graph based card embeddings for all cards retrieved from the transaction database. 
     
     
         7 . The method as claimed in  claim 6 , further comprising:
 identifying a plurality of risky cards having performed a higher number of restricted MCC transactions compared to a predefined restricted MCC threshold value; and   generating the graph based risky card embedding for training the neural network model by computing an average of a plurality of graph based risky card embeddings generated using the HGCN model, the plurality of graph based risky card embeddings associated with the plurality of risky cards.   
     
     
         8 . The method as claimed in  claim 1 , wherein a respective cosine similarity function is used for computing the sequential merchant similarity value, the graph based merchant similarity value, the graph based card similarity value, the sequential card similarity value and the sequential merchant MCC similarity value. 
     
     
         9 . The method as claimed in  claim 1 , wherein the plurality of features further comprises a transaction amount, a transaction status, a transaction time, a transaction type, a number of restricted MCC transactions processed using the payment instrument, and an approval rate for the number of restricted MCC transactions processed using the payment instrument. 
     
     
         10 . The method as claimed in  claim 1 , wherein the neural network model trained to predict the plurality of reconstructed transaction windows from the plurality of transaction windows is a Long Short-Term Memory (LSTM) Autoencoder with an attention layer. 
     
     
         11 . The method as claimed in  claim 1 , wherein a mean squared error between a plurality of features of a transaction and a plurality of reconstructed features of a reconstructed transaction is calculated to compute a corresponding reconstructed loss value for the transaction. 
     
     
         12 . The method as claimed in  claim 1 , wherein the threshold value is previously determined based on a plurality of validation phase reconstruction loss values computed after training of the neural network model during a validation phase, a validation dataset utilized for computing the plurality of validation phase reconstruction loss values comprising a plurality of transaction windows comprising at least one re-routing transaction window, wherein the at least one re-routing transaction window further comprises at least one declined transaction under restricted MCC and a corresponding re-routed transaction. 
     
     
         13 . The method as claimed in  claim 1 , wherein each transaction window of the plurality of transaction windows comprises first transaction as the transaction declined under the restricted MCC. 
     
     
         14 . A system comprising:
 a memory comprising executable instructions; and   a processor communicably coupled to a communication interface and configured to execute the instructions to cause the system to at least:
 retrieve a plurality of transaction windows from a transaction database, each transaction window of the plurality of transaction windows comprising a transaction declined under a restricted Merchant Category Code (MCC) and each transaction window associated with a respective payment instrument; 
 access a plurality of features associated with each transaction of each transaction window from the transaction database, the plurality of features comprising a sequential merchant similarity value with respect to a previously generated sequential restricted MCC merchant embedding, a graph based merchant similarity value with respect to a previously generated graph based restricted MCC merchant embedding, a sequential card similarity value with respect to a previously generated sequential risky card embedding, a graph based card similarity value with respect to a previously generated graph based risky card embedding, and a sequential merchant MCC similarity value with respect to a previously generated sequential restricted MCC embedding; 
 generate an input dataset by appending the plurality of features to each associated transaction for each transaction of each transaction window of the plurality of transaction windows; 
 predict an output dataset comprising a plurality of reconstructed transaction windows based on feeding the input dataset to a trained neural network model; 
 compute a corresponding reconstruction loss value for each transaction of each transaction window wherein a reconstruction loss value for an individual transaction is computed by comparing the individual transaction with a corresponding reconstructed transaction; 
 compare the corresponding reconstruction loss value for each transaction of each transaction window with a pre-determined threshold value; and 
 if a corresponding reconstruction loss value for a transaction of a transaction window is higher than the pre-determined threshold value, identify the transaction as a re-routed transaction corresponding to the transaction declined under the restricted MCC of the transaction window. 
   
     
     
         15 . The system as claimed in  claim 14 , wherein the restricted Merchant Category Code (MCC) is MCC 7995. 
     
     
         16 . The system as claimed in  claim 14 , wherein the plurality of features further comprises a transaction amount, a transaction status, a transaction time, a transaction type, a number of restricted MCC transactions processed using the payment instrument, and an approval rate for the number of restricted MCC transactions processed using the payment instrument. 
     
     
         17 . The system as claimed in  claim 14 , wherein a mean squared error between a plurality of features of a transaction and a plurality of reconstructed features of a reconstructed transaction is calculated to compute a corresponding reconstructed loss value for the transaction. 
     
     
         18 . The system as claimed in  claim 14 , wherein the threshold value is previously determined based on a plurality of validation phase reconstruction loss values computed after training of the neural network model during a validation phase, a validation dataset utilized for computing the plurality of validation phase reconstruction loss values comprising a plurality of transaction windows comprising at least one re-routing transaction window, wherein the at least one re-routing transaction window further comprises at least one declined transaction under restricted MCC and a corresponding re-routed transaction. 
     
     
         19 . A computer-implemented method comprising:
 retrieving, by a processor, a transaction window associated with a payment card from a transaction database, first transaction of the transaction window being a transaction declined under a restricted Merchant Category Code (MCC);   accessing, by the processor, a plurality of transactional features, a plurality of card features, and a plurality of merchant features associated with each transaction of the transaction window from the transaction database, the plurality of merchant features comprising a sequential merchant similarity value with respect to a previously generated sequential restricted MCC merchant embedding, a graph based merchant similarity value with respect to a previously generated graph based restricted MCC merchant embedding, and a sequential merchant MCC similarity value with respect to a previously generated sequential restricted MCC embedding and the plurality of card features comprising a sequential card similarity value with respect to a previously generated sequential risky card embedding, and a graph based card similarity value with respect to a previously generated graph based risky card embedding;   generating, by the processor, an input dataset by appending the plurality of transactional features, the plurality of card features, and the plurality of merchant features to each associated transaction for each transaction of the transaction window;   predicting, by the processor, an output dataset comprising a reconstructed transaction window based on feeding the input dataset to a previously trained Long Short-Term Memory (LSTM) Autoencoder with an attention layer model;   computing, by the processor, a corresponding reconstruction loss value for each transaction of the transaction window by comparing each transaction of the input dataset with each corresponding reconstructed transaction of the reconstructed transaction window of the output dataset;   comparing, by the processor, the corresponding reconstruction loss value for each transaction of the transaction window with a pre-determined threshold value, wherein the threshold value is previously determined based on a plurality of validation phase reconstruction loss values computed after training of the LSTM autoencoder with attention layer model during a validation phase, a validation dataset utilized for computing the plurality of validation phase reconstruction loss values comprising a plurality of transaction windows comprising at least one re-routing transaction window, the at least one re-routing transaction window further comprising at least one declined transaction under the restricted MCC and a corresponding re-routed transaction; and   if a corresponding reconstruction loss value for a transaction is higher than the pre-determined threshold value, identifying, by the processor, the transaction as a re-routed transaction corresponding to the transaction declined under the restricted MCC.   
     
     
         20 . The method as claimed in  claim 19 , wherein the plurality of transactional features further comprises a transaction amount, a transaction status, a transaction time, a transaction type, and wherein the plurality of card features comprises a number of restricted MCC transactions processed using the payment card, and an approval rate for the number of restricted MCC transactions processed using the payment card.